323 lines
15 KiB
C++
323 lines
15 KiB
C++
/*******************************************************************************
|
|
* Copyright (c) 2015-2018 Skymind, Inc.
|
|
*
|
|
* This program and the accompanying materials are made available under the
|
|
* terms of the Apache License, Version 2.0 which is available at
|
|
* https://www.apache.org/licenses/LICENSE-2.0.
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
|
|
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
|
|
* License for the specific language governing permissions and limitations
|
|
* under the License.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
******************************************************************************/
|
|
|
|
#ifndef CUDA_LAMBDA_HELPER
|
|
#define CUDA_LAMBDA_HELPER
|
|
|
|
#include <pointercast.h>
|
|
#include <op_boilerplate.h>
|
|
#include <helpers/shape.h>
|
|
#include <cuda.h>
|
|
#include <cuda_runtime.h>
|
|
|
|
static Nd4jLong __device__ __noinline__ __getIndexOffset(Nd4jLong index, Nd4jLong *shapeInfo, Nd4jLong length) {
|
|
return shape::getIndexOffset(index, shapeInfo, length);
|
|
}
|
|
|
|
static Nd4jLong __device__ __noinline__ __length(Nd4jLong *shapeInfo) {
|
|
return shape::length(shapeInfo);
|
|
}
|
|
|
|
template <typename T, typename Lambda> static _CUDA_G void lambdaKernel(void* vx, Nd4jLong *xShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda);
|
|
template <typename T, typename Lambda> static _CUDA_G void lambdaIndexedKernel(void* vx, Nd4jLong *xShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda);
|
|
template <typename T, typename Lambda> static _CUDA_G void lambdaIndexedPairwiseKernel(void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda);
|
|
template <typename T, typename Lambda> static _CUDA_G void lambdaPairwiseKernel(void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda);
|
|
template <typename T, typename Lambda> static _CUDA_G void lambdaTriplewiseKernel(void* vw, Nd4jLong *wShapeInfo, void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda);
|
|
|
|
template <typename T>
|
|
class LambdaHelper {
|
|
public:
|
|
|
|
template <typename Lambda>
|
|
FORCEINLINE static void lambdaLauncher(cudaStream_t *stream, void* vx, Nd4jLong *xShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
lambdaKernel<T, Lambda><<<256, 512, 1024, *stream>>>(vx, xShapeInfo, vz, zShapeInfo, lambda);
|
|
auto err = cudaStreamSynchronize(*stream);
|
|
if (err != 0)
|
|
throw std::runtime_error("NDArray::applyLambda execution failed");
|
|
}
|
|
|
|
template <typename Lambda>
|
|
FORCEINLINE static void lambdaIndexedLauncher(cudaStream_t *stream, void* vx, Nd4jLong *xShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
lambdaIndexedKernel<T, Lambda><<<256, 512, 1024, *stream>>>(vx, xShapeInfo, vz, zShapeInfo, lambda);
|
|
auto err = cudaStreamSynchronize(*stream);
|
|
if (err != 0)
|
|
throw std::runtime_error("NDArray::applyIndexedLambda execution failed");
|
|
}
|
|
|
|
template <typename Lambda>
|
|
FORCEINLINE static void lambdaPairwiseLauncher(cudaStream_t *stream, void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
lambdaPairwiseKernel<T, Lambda><<<256, 512, 1024, *stream>>>(vx, xShapeInfo, vy, yShapeInfo, vz, zShapeInfo, lambda);
|
|
auto err = cudaStreamSynchronize(*stream);
|
|
if (err != 0)
|
|
throw std::runtime_error("NDArray::applyPairwiseLambda execution failed");
|
|
}
|
|
|
|
template <typename Lambda>
|
|
FORCEINLINE static void lambdaIndexedPairwiseLauncher(cudaStream_t *stream, void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
lambdaIndexedPairwiseKernel<T, Lambda><<<256, 512, 1024, *stream>>>(vx, xShapeInfo, vy, yShapeInfo, vz, zShapeInfo, lambda);
|
|
auto err = cudaStreamSynchronize(*stream);
|
|
if (err != 0)
|
|
throw std::runtime_error("NDArray::applyIndexedPairwiseLambda execution failed");
|
|
}
|
|
|
|
template <typename Lambda>
|
|
FORCEINLINE static void lambdaTriplewiseLauncher(cudaStream_t *stream, void* vw, Nd4jLong *wShapeInfo, void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
lambdaTriplewiseKernel<T, Lambda><<<256, 512, 1024, *stream>>>(vw, wShapeInfo, vx, xShapeInfo, vy, yShapeInfo, vz, zShapeInfo, lambda);
|
|
auto err = cudaStreamSynchronize(*stream);
|
|
if (err != 0)
|
|
throw std::runtime_error("NDArray::applyTriplewiseLambda execution failed");
|
|
}
|
|
};
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename T, typename Lambda>
|
|
static _CUDA_G void lambdaKernel(void* vx, Nd4jLong *xShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
auto x = reinterpret_cast<T*>(vx);
|
|
auto z = reinterpret_cast<T*>(vz);
|
|
|
|
auto xEws = shape::elementWiseStride(xShapeInfo);
|
|
auto zEws = shape::elementWiseStride(zShapeInfo);
|
|
|
|
auto xOrder = shape::order(xShapeInfo);
|
|
auto zOrder = shape::order(zShapeInfo);
|
|
|
|
auto zLength = __length(zShapeInfo);
|
|
|
|
auto tid = threadIdx.x + blockIdx.x * blockDim.x;
|
|
|
|
if (xEws >= 1 && zEws >= 1 && xOrder == zOrder) {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x)
|
|
z[e * zEws] = lambda(x[e * xEws]);
|
|
} else {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x) {
|
|
auto xOffset = __getIndexOffset(e, xShapeInfo, zLength);
|
|
auto zOffset = __getIndexOffset(e, zShapeInfo, zLength);
|
|
|
|
z[zOffset] = lambda(x[xOffset]);
|
|
}
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename T, typename Lambda>
|
|
static _CUDA_G void lambdaIndexedKernel(void* vx, Nd4jLong *xShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
auto x = reinterpret_cast<T*>(vx);
|
|
auto z = reinterpret_cast<T*>(vz);
|
|
|
|
auto xEws = shape::elementWiseStride(xShapeInfo);
|
|
auto zEws = shape::elementWiseStride(zShapeInfo);
|
|
|
|
auto xOrder = shape::order(xShapeInfo);
|
|
auto zOrder = shape::order(zShapeInfo);
|
|
|
|
auto zLength = __length(zShapeInfo);
|
|
|
|
auto tid = threadIdx.x + blockIdx.x * blockDim.x;
|
|
|
|
if (xEws >= 1 && zEws >= 1 && xOrder == zOrder) {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x)
|
|
z[e * zEws] = lambda(e, x[e * xEws]);
|
|
} else {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x) {
|
|
auto xOffset = __getIndexOffset(e, xShapeInfo, zLength);
|
|
auto zOffset = __getIndexOffset(e, zShapeInfo, zLength);
|
|
|
|
z[zOffset] = lambda(e, x[xOffset]);
|
|
}
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename T, typename Lambda>
|
|
static _CUDA_G void lambdaIndexedPairwiseKernel(void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
auto x = reinterpret_cast<T*>(vx);
|
|
auto y = reinterpret_cast<T*>(vy);
|
|
auto z = reinterpret_cast<T*>(vz);
|
|
|
|
auto xEws = shape::elementWiseStride(xShapeInfo);
|
|
auto yEws = shape::elementWiseStride(yShapeInfo);
|
|
auto zEws = shape::elementWiseStride(zShapeInfo);
|
|
|
|
auto xOrder = shape::order(xShapeInfo);
|
|
auto yOrder = shape::order(yShapeInfo);
|
|
auto zOrder = shape::order(zShapeInfo);
|
|
|
|
auto zLength = __length(zShapeInfo);
|
|
|
|
auto tid = threadIdx.x + blockIdx.x * blockDim.x;
|
|
|
|
if (xEws >= 1 && yEws >= 1 && zEws >= 1 && xOrder == zOrder && yOrder == xOrder) {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x)
|
|
z[e * zEws] = lambda(e, x[e * xEws], y[e * yEws]);
|
|
} else {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x) {
|
|
auto xOffset = __getIndexOffset(e, xShapeInfo, zLength);
|
|
auto yOffset = __getIndexOffset(e, yShapeInfo, zLength);
|
|
auto zOffset = __getIndexOffset(e, zShapeInfo, zLength);
|
|
|
|
z[zOffset] = lambda(e, x[xOffset], y[yOffset]);
|
|
}
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename T, typename Lambda>
|
|
static _CUDA_G void lambdaPairwiseKernel(void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
auto x = reinterpret_cast<T*>(vx);
|
|
auto y = reinterpret_cast<T*>(vy);
|
|
auto z = reinterpret_cast<T*>(vz);
|
|
|
|
auto xEws = shape::elementWiseStride(xShapeInfo);
|
|
auto yEws = shape::elementWiseStride(yShapeInfo);
|
|
auto zEws = shape::elementWiseStride(zShapeInfo);
|
|
|
|
auto xOrder = shape::order(xShapeInfo);
|
|
auto yOrder = shape::order(yShapeInfo);
|
|
auto zOrder = shape::order(zShapeInfo);
|
|
|
|
auto zLength = __length(zShapeInfo);
|
|
|
|
auto tid = threadIdx.x + blockIdx.x * blockDim.x;
|
|
|
|
if (xEws >= 1 && yEws >= 1 && zEws >= 1 && xOrder == zOrder && yOrder == xOrder) {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x)
|
|
z[e * zEws] = lambda(x[e * xEws], y[e * yEws]);
|
|
} else {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x) {
|
|
auto xOffset = __getIndexOffset(e, xShapeInfo, zLength);
|
|
auto yOffset = __getIndexOffset(e, yShapeInfo, zLength);
|
|
auto zOffset = __getIndexOffset(e, zShapeInfo, zLength);
|
|
|
|
z[zOffset] = lambda(x[xOffset], y[yOffset]);
|
|
}
|
|
}
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////
|
|
template <typename T, typename Lambda>
|
|
static _CUDA_G void lambdaTriplewiseKernel(void* vw, Nd4jLong *wShapeInfo, void* vx, Nd4jLong *xShapeInfo, void* vy, Nd4jLong *yShapeInfo, void *vz, Nd4jLong *zShapeInfo, Lambda lambda) {
|
|
auto w = reinterpret_cast<T*>(vw);
|
|
auto x = reinterpret_cast<T*>(vx);
|
|
auto y = reinterpret_cast<T*>(vy);
|
|
auto z = reinterpret_cast<T*>(vz);
|
|
|
|
auto wEws = shape::elementWiseStride(wShapeInfo);
|
|
auto xEws = shape::elementWiseStride(xShapeInfo);
|
|
auto yEws = shape::elementWiseStride(yShapeInfo);
|
|
auto zEws = shape::elementWiseStride(zShapeInfo);
|
|
|
|
auto wOrder = shape::order(wShapeInfo);
|
|
auto xOrder = shape::order(xShapeInfo);
|
|
auto yOrder = shape::order(yShapeInfo);
|
|
auto zOrder = shape::order(zShapeInfo);
|
|
|
|
auto zLength = __length(zShapeInfo);
|
|
|
|
auto tid = threadIdx.x + blockIdx.x * blockDim.x;
|
|
|
|
if (wEws > 1 && xEws >= 1 && yEws >= 1 && zEws >= 1 && xOrder == zOrder && yOrder == xOrder && wOrder == xOrder) {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x)
|
|
z[e * zEws] = lambda(w[e * wEws], x[e * xEws], y[e * yEws]);
|
|
} else {
|
|
for (uint e = tid; e < zLength; e += blockDim.x * gridDim.x) {
|
|
auto wOffset = __getIndexOffset(e, wShapeInfo, zLength);
|
|
auto xOffset = __getIndexOffset(e, xShapeInfo, zLength);
|
|
auto yOffset = __getIndexOffset(e, yShapeInfo, zLength);
|
|
auto zOffset = __getIndexOffset(e, zShapeInfo, zLength);
|
|
|
|
z[zOffset] = lambda(w[wOffset], x[xOffset], y[yOffset]);
|
|
}
|
|
}
|
|
}
|
|
|
|
#endif
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
template<typename Lambda>
|
|
void NDArray::applyLambda(Lambda func, NDArray* target) {
|
|
auto result = target == nullptr ? this : target;
|
|
auto dtype = this->dataType();
|
|
|
|
if (dtype != result->dataType())
|
|
throw std::runtime_error("NDArray::applyLambda X/Z data types must be the same");
|
|
//throw datatype_exception::build("NDArray::applyLambda X/Z data types must be the same", dtype, result->dataType());
|
|
prepareSpecialUse({result}, {this});
|
|
BUILD_SINGLE_SELECTOR(dtype, LambdaHelper ,::lambdaLauncher(this->_context->getCudaStream(), this->specialBuffer(), this->specialShapeInfo(), result->specialBuffer(), result->specialShapeInfo(), func), LIBND4J_TYPES);
|
|
registerSpecialUse({result}, {this});
|
|
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
template<typename Lambda>
|
|
void NDArray::applyPairwiseLambda(const NDArray* other, Lambda func, NDArray* target) {
|
|
auto result = target == nullptr ? this : target;
|
|
auto dtype = this->dataType();
|
|
|
|
if (dtype != result->dataType() || dtype != other->dataType())
|
|
throw std::runtime_error("NDArray::applyPairwiseLambda X/Y/Z data types must be the same");
|
|
//throw datatype_exception::build("NDArray::applyLambda X/Z data types must be the same", dtype, result->dataType());
|
|
|
|
prepareSpecialUse({result}, {this, other});
|
|
BUILD_SINGLE_SELECTOR(dtype, LambdaHelper ,::lambdaPairwiseLauncher(this->_context->getCudaStream(), this->specialBuffer(), this->specialShapeInfo(), other->getSpecialBuffer(), other->getSpecialShapeInfo(), result->specialBuffer(), result->specialShapeInfo(), func), LIBND4J_TYPES);
|
|
registerSpecialUse({result}, {this, other});
|
|
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
template <typename Lambda>
|
|
void NDArray::applyIndexedLambda(Lambda func, NDArray* target) {
|
|
auto result = target == nullptr ? this : target;
|
|
auto dtype = this->dataType();
|
|
if (dtype != result->dataType())
|
|
throw std::runtime_error("NDArray::applyIndexedLambda X/Z data types must be the same");
|
|
|
|
prepareSpecialUse({result}, {this});
|
|
BUILD_SINGLE_SELECTOR(dtype, LambdaHelper ,::lambdaIndexedLauncher(this->_context->getCudaStream(), this->specialBuffer(), this->specialShapeInfo(), result->specialBuffer(), result->specialShapeInfo(), func), LIBND4J_TYPES);
|
|
registerSpecialUse({result}, {this});
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
template <typename Lambda>
|
|
void NDArray::applyIndexedPairwiseLambda(NDArray* other, Lambda func, NDArray* target) {
|
|
auto result = target == nullptr ? this : target;
|
|
auto dtype = this->dataType();
|
|
if (dtype != result->dataType() || dtype != other->dataType())
|
|
throw std::runtime_error("NDArray::applyIndexedPairwiseLambda X/Y/Z data types must be the same");
|
|
|
|
prepareSpecialUse({result}, {this, other});
|
|
BUILD_SINGLE_SELECTOR(dtype, LambdaHelper ,::lambdaIndexedPairwiseLauncher(this->_context->getCudaStream(), this->specialBuffer(), this->specialShapeInfo(), other->getSpecialBuffer(), other->getSpecialShapeInfo(), result->specialBuffer(), result->specialShapeInfo(), func), LIBND4J_TYPES);
|
|
registerSpecialUse({result}, {this, other});
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
template <typename Lambda>
|
|
void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, Lambda func, NDArray* target) {
|
|
auto result = target == nullptr ? this : target;
|
|
auto dtype = this->dataType();
|
|
|
|
if (dtype != result->dataType() || dtype != second->dataType() || dtype != third->dataType())
|
|
throw std::runtime_error("NDArray::applyTriplewiseLambda X/Y/Z data types must be the same");
|
|
|
|
prepareSpecialUse({result}, {this, second, third});
|
|
BUILD_SINGLE_SELECTOR(dtype, LambdaHelper ,::lambdaTriplewiseLauncher(this->_context->getCudaStream(), this->specialBuffer(), this->specialShapeInfo(), second->specialBuffer(), second->specialShapeInfo(), third->specialBuffer(), third->specialShapeInfo(), result->specialBuffer(), result->specialShapeInfo(), func), LIBND4J_TYPES);
|
|
registerSpecialUse({result}, {this, second, third});
|
|
}
|
|
|
|
|
|
|
|
|
|
|